Actors

The dictator interacts with a military agent and also confronts a mass outsider movement.

Military. The two possible types of military in the model are competent and personalist. Each type encompasses and condenses numerous strategic actions that real-world dictators can take to organize their coercive apparatus. These include how to select officers and rank-and-file soldiers for the conventional military; how much information flow to allow across units, which affects the unitary versus fragmented nature of the security apparatus; and whether (and how) to create or maintain paramilitary units and secret police (among recent work, see Blaydes Reference Blaydes2018; De Bruin Reference De Bruin2020; Geddes, Wright, and Frantz Reference Geddes, Wright and Frantz2018; Greitens Reference Greitens2016; Harkness Reference Harkness2016; Lyall Reference Lyall2020; Talmadge Reference Talmadge2015). In a typical competent military, the ruler pursues socially inclusive recruitment strategies for the officer corps and rank-and-file soldiers in the military and creates a professional apparatus distinguished by meritocratic promotion and a disciplined hierarchical command. For example, upon attaining power in 1995, the Tutsi-dominated Rwandan Patriotic Front “sought to ensure the security and defense of the country by forming a coherent national defense force.” They did so by incorporating numerous Hutu soldiers from the previous regime, which facilitated “one of the most capable militaries in Africa” (Burgess Reference Burgess and Licklider2014, 92, 97). By contrast, dictators can prioritize personalist ties by creating socially exclusive militaries in which they stack the officer corps with unqualified family members, coethnics, and groups with a weak domestic power base. They can complement socially exclusive recruitment with safeguards such as fractured communication between officers and additional paramilitary units. These were hallmarks of Saddam Hussein’s rule in Iraq, in particular by the 1990s (Blaydes Reference Blaydes2018; Quinlivan Reference Quinlivan1999).

Throughout, I primarily refer to the coercive agent with whom the ruler interacts as “the military.” Despite distinct organizations within the overall coercive apparatus, high-ranking officers in the conventional army typically control the fate of the regime when confronting a major insurgency or mass urban protests. The conventional military is also crucial for confronting foreign threats (Finer Reference Finer1997; Talmadge Reference Talmadge2015), which I address in the conclusion. By contrast, rulers typically rely on the police and specialized internal security agencies for everyday repression techniques (Greitens Reference Greitens2016).

Highlighting the importance of the conventional military against mass domestic threats, Svolik (Reference Svolik2013, 765) argues, “When underlying, polity-wide conflict results in threats to the regime that take the particular form of mass, organized, and potentially violent opposition, the military is the only force capable of defeating them” [emphasis added]. Reflecting on events during the Arab Spring, Bellin (Reference Bellin2012, 130–1) argues that “when it comes to mass unrest such as that seen on Habib Bourguiba Avenue or Tahrir Square, where tens of thousands of angry people assembled to demand an end to the regime in power, such mobilization usually overwhelms the capacity of the regular police and/or intelligence services. In that case, regime survival turns on the military (primarily the army) … to contain a mass uprising” [emphasis added].

The importance of the state military is readily apparent when confronting armed insurgent groups. Regarding urban protests, Table 1 summarizes data from Brancati (Reference Brancati2016) on pro-democracy protests between 1989 and 2011. Rulers called on the police in approximately 60% of protests, regardless of the size of the protest. By contrast, dictators typically called on the military only when protests were quite large (greater than 100,000 participants), usually after the police failed to quell the movement.

Table 1. Coercive Responses to Pro-Democracy Protests, 1989–2011

Note: Table 1 presents the percentage of cases in which the regime deployed each coercive apparatus, and the columns distinguish the number of people that participated in the protests.

Mass outsider threat. Mass threats consist of any groups of people outside the ruling coalition. This includes members of ethnic groups that lack positions in the central government, rebel groups, societal organizations including labor unions and religious groups, students and unemployed youth, and rural peasants. These groups contrast with insiders such as the ruler, their inner circle, and high-ranking military officials. Following the model analysis, I provide numerous examples of outsider movements.

The Core Tension: Coercive Capabilities versus Posttransition Fate

For the dictator, the core tension is that the same traits that make a competent military more coercively capable of defeating outsider movements also enhance their likelihood of remaining intact following a regime transition, which I refer to as their posttransition fate.

Coercive capabilities. Recruiting and promoting broadly among social groups, in particular for officer roles, boosts the coercive capabilities of the military by enabling more talented soldiers to achieve high-ranking positions. Relatedly, unifying the command structure facilitates coordinated operations and communication that can handle “multi-city riot control, counterinsurgency, or other widespread forms of popular unrest” (Greitens Reference Greitens2016, 31).

Numerous scholars emphasize the converse drawbacks of personalist-oriented and coup-proofed militaries. Promoting officers on grounds of ethnic affinity rather than merit hinders battlefield performance, as does impeding communication across units to reduce opportunities for coup attempts (Talmadge Reference Talmadge2015) or subordinating certain rank-and-file soldiers based on ethnicity (Lyall Reference Lyall2020). Narrow and ethnically biased recruitment strategies can create manpower deficits (Quinlivan Reference Quinlivan1999) and undermine intelligence networks and counterinsurgency capabilities in areas populated by excluded groups (Roessler Reference Roessler2016). Information deficits also impede precise targeting of repression. Indiscriminate repression yields a higher probability of triggering the “repression–dissent” paradox whereby repression spurs rather than quells societal mobilization (Ritter and Conrad Reference Ritter and Conrad2016).

Posttransition fate. Yet personalist militaries also offer benefits for the ruler. The livelihoods of their members are typically intertwined with the survival of the incumbent regime, in particular because of restrictive social recruitment. These units are likely to be heavily purged or outright disbanded if the incumbent loses office regardless of the nature of the next ruling group. In some cases, the military is primarily composed of members of the ruler’s ethnic (or other politically relevant identity) group. If they are members of a minority group that is unlikely to continue to control the government upon the incumbent regime falling, they anticipate a poor posttransition fate. This implication is similar for militaries primarily composed of members of “weak” groups that lack a domestic power base.

As an example of coethnics, van Dam (Reference van Dam2011, 134–5) commented on the perils of Syria’s minority-dominated regime just prior to the Arab Spring movement: “it is very difficult to imagine a scenario in which the present narrowly based, totalitarian regime, dominated by members of the Alawi minority, who traditionally have been discriminated against by the Sunni majority” could count on “much understanding from a … regime which would for instance be dominated by members of the Sunni majority.” This statement applied equally to the Alawi–dominated Syrian military (Quinlivan Reference Quinlivan1999). As an example of non-coethnics with a weak domestic power base, Finer (Reference Finer1997, 301) discusses how many historical dictators employed eunuchs in high-ranking positions because they were more “faithful than most men … eunuchs were despised by the rest of mankind, hence they were dependent on a patron for protection.”

By contrast, competent and socially inclusive militaries face better prospects for surviving largely intact following a regime transition. Reflecting on examples of professional militaries in Latin America in the 1980s, Geddes (Reference Geddes1999, 131) claims, “For officers, there is life after democracy, as all but the highest regime officials can usually return to the barracks with their status and careers untarnished.” Bellin (Reference Bellin2012, 133) argues that elites in an institutionalized, as opposed to patrimonially organized, military “will have a distinct mission identity, and career path. Under these conditions the military elite will be able to imagine separation from the regime and life beyond the regime.” She proposes that this factor helps to explain why the military defected in Egypt and Tunisia during the Arab Spring.

However, competent militaries do not always anticipate a favorable posttransition fate upon handing power to outsiders. In the model, I allow this factor to vary. After the model analysis, I discuss how mass organizations that seek radical distribution away from the ruling group (including Marxists, violence-espousing Islamists, and ethnically organized rebels) create a poor posttransition fate. The outcome for competent militaries also depends on their actions under the incumbent regime. If they repress the masses, either when trying to save the incumbent regime or when attempting to establish a military dictatorship, they fear retribution for human rights abuses if their repression fails and the outsider movement gains power. This assumption reflects research on transitional justice and the agency problems underlying repression (Nalepa Reference Nalepa2010; Tyson Reference Tyson2018).

In the model, the military’s posttransition consumption directly affects their motives to defect (in the analysis, I explain why their posttransition fate also affects their incentives to stage a coup). When facing major urban protests, military defection typically takes the form of refusing to shoot, as in the aforementioned cases of Egypt and Tunisia. When facing armed insurgents, defection entails soldiers fleeing or joining the other side. For example, in Chad in 1990, the Patriotic Salvation Movement (MPS) rebel group faced a manpower disadvantage of 2,000 soldiers compared with the 30,000-strong state military. Yet the rebels defeated the government when soldiers from the state military “fle[d] or defect[ed] to the MPS” (Dixon and Sarkees Reference Dixon and Sarkees2015, 643). Consequently, “the new government was generally welcomed. In N’Djamena many former ministers and party officials rallied to the new government” (Nolutshungu Reference Nolutshungu1996, 246).Footnote ⁶

Contributions to Existing Research

My model draws from disparate strands of the literature. In the introduction, I discussed arguments from numerous recent articles that, collectively, constitute the conventional characterization of the guardianship dilemma.Footnote ⁷ I depart from these theories by incorporating a strategic option for the military to defect when facing an outsider threat in addition to the standard disloyalty option of launching a coup. Introducing a defection option forces us to think about what happens next for the military if the incumbent regime falls, as well as their motivation to prevent this outcome from occurring.

My approach builds in part off McMahon and Slantchev’s (Reference McMahon and Slantchev2015) insightful critique of the guardianship dilemma. In particular, they critique the conventional implication that stronger outsider threats make a ruler more reliant on the military and thus more susceptible to coups. They demonstrate that stronger outsider threats make the military more fearful of staging a coup: if they succeed, they still have to confront the strong outsider. I incorporate their core premise that the military thinks about what comes next if their coup attempt succeeds, which contrasts with the standard and less satisfying assumption that the outsider threat simply disappears if the military takes over. However, my model differs in two significant ways that generate my new findings. First, like other models of the guardianship dilemma, McMahon and Slantchev do not include a strategic option for the military to defect. Second, they assume that any military fares badly if a regime transition occurs. By contrast, by including a defection option and allowing the competent military’s posttransition fate to vary, my model recovers important implications from both the conventional logic and McMahon and Slantchev’s model as special cases.Footnote ⁸

Other formal models illuminate the agency problem of military defection by analyzing the commitment problem inherent in paying security agents (Myerson Reference Myerson2008; Tyson Reference Tyson2018). However, they do not analyze what type of military agent the ruler prefers, nor do they incorporate a coup option. Other authors discuss important attributes of military composition such as loyalty, efficiency, and cost (Finer Reference Finer1997); will and capacity (Bellin Reference Bellin2012); and cohesion and scope (Levitsky and Way Reference Levitsky and Way2010). I build on these conceptual innovations to develop strategic microfoundations for choices by the ruler and coercive agent in the context of the broader guardianship dilemma.

I also take a new approach relative to the few formal-theoretic articles on the loyalty–competence trade-off in dictatorships. The present idea that competent militaries have a better posttransition fate relates to Zakharov’s (Reference Zakharov2016) assumption that high-quality viziers have a better outside option if they defect from the incumbent. My approach differs by engaging with two core elements of the logic of the guardianship dilemma. First, I allow the coercive agent the strategic option of staging a coup. Second, I take comparative statics on the strength of outsider threats, which affects both (a) the military’s optimal choice among their strategic options and (b) the dictator’s optimal choice of agent. I also depart from Egorov and Sonin (Reference Egorov and Sonin2011) by assuming that competent and personalist militaries differ in their inherent ability to save the regime (conditional on acting loyally). By contrast, in their model, agents differ in the precision of their information about the nature of the outsider threat. Yet regardless of their informational competence, acting loyally is sufficient to save the regime.

Sequence of Moves

Two strategic players, a dictator and a military agent, make sequential choices in a one-shot game. They collectively encounter a mass outsider threat (represented by a Nature move) endowed with coercive strength $ {\theta}_{\mathrm{out}}>0 $ .

The dictator cares only about survival in office, consuming 1 upon survival (i.e., if the military acts loyally and this repressive effort succeeds) and 0 otherwise. The dictator moves first and chooses to construct either a competent military endowed with coercive strength $ {\theta}_{\mathrm{comp}}>0 $ or a personalist military endowed with coercive strength $ {\theta}_{\mathrm{pers}}>0 $ . When referring generically to the military’s coercive endowment, I write $ {\theta}_{\mathrm{mil}} $ .

After this move, the chosen military agent learns how much they will consume if the status quo regime survives. Nature draws π _sq ~ F = U[0, $ {\pi}_{\mathrm{sq}}^{\mathrm{max}}\Big] $ , for strictly positive and large $ {\pi}_{\mathrm{sq}}^{\mathrm{max}} $ .Footnote ⁹ Because the ruler knows only the prior distribution of this draw when moving, this Nature move is reduced form for a bargaining interaction in which the ruler faces some friction to compensating the military, such as a commitment problem (Acemoglu, Ticchi, and Vindigni Reference Acemoglu, Ticchi and Vindigni2010) or a contracting problem (Svolik Reference Svolik2013).

The second, and final, strategic move is that the military agent selects among three options. They can exhibit loyalty by using repression to try to save the regime. Alternatively, they can exercise either of two disloyalty options: defecting against the mass outsider threat or staging a coup.

Defection is the most straightforward to describe. Choosing this option ensures that the incumbent regime falls and a transition occurs to a regime governed by leaders of the mass outsider movement. Under this outcome, the competent military consumes $ {\pi}_{\mathrm{trans}}>0 $ , which parameterizes their posttransition fate. I impose a finite upper bound on $ {\pi}_{\mathrm{trans}} $ that I explain in more detail below. The personalist military consumes 0 if outsider takeover occurs by any means, including after defection.

If the military acts loyally, then the regime survives the mass threat with probability $ p\left({\theta}_{\mathrm{mil}},{\theta}_{\mathrm{out}}\right)\in \left(0,1\right) $ .Footnote ¹⁰ For either military actor, regime survival yields consumption of $ {\pi}_{\mathrm{sq}} $ . I often refer to the competent military’s probability of defeating the outsider as $ {p}_{\mathrm{comp}}\equiv p\left({\theta}_{\mathrm{comp}},{\theta}_{\mathrm{out}}\right) $ and the personalist military’s probability as $ {p}_{\mathrm{pers}}\equiv p\left({\theta}_{\mathrm{pers}},{\theta}_{\mathrm{out}}\right) $ . With complementary probability, repression fails and the regime falls. Failed repression yields consumption of $ \gamma \cdot {\pi}_{\mathrm{trans}} $ for the competent military, with $ \gamma \in \left(0,1\right) $ , which reflects punishment for committing human rights abuses.Footnote ¹¹ The personalist military consumes 0 following failed repression.

Finally, I assume that a coup attempt displaces the incumbent ruler for sure, but the military may fail to consolidate power. Specifically, a coup succeeds at establishing a military dictatorship with probability $ \alpha \left({\theta}_{\mathrm{out}}\right)\cdot p\left({\theta}_{\mathrm{mil}},{\theta}_{\mathrm{out}}\right) $ . This is correlated with the probability of preventing outsider rule upon acting loyally, but is strictly lower because I assume that coups destabilize the center, $ \alpha \left({\theta}_{\mathrm{out}}\right)\in \left(0,1\right) $ for all $ {\theta}_{\mathrm{out}}\ge 0 $ . Consequently, the military is less likely to defeat the outsider threat following a coup attempt than upon acting loyally. I incorporate the additional natural premise that stronger outsiders are better able to exploit voids at the center by assuming $ \frac{\partial \alpha }{\partial {\theta}_{\mathrm{out}}}<0 $ . Establishing a military dictatorship yields consumption of 1 for either type of military. With complementary probability following a coup, the military fails to cling to power and a regime transition to mass rule occurs. This yields the same payoffs as when the military acts loyally but fails to save the regime: $ \gamma \cdot {\pi}_{\mathrm{trans}} $ for the competent military and 0 for the personalist military.Footnote ¹²

Figure 1 presents the game tree, in which the last Nature node reflects the “action” by the unmodeled masses actor. Table 2 summarizes every parameter and choice variable.

Table 2. Summary of Parameters and Choice Variables

Figure 1. Game Tree

Formalizing the Core Tension

For the dictator, the core tension is that the same traits that make a competent military more coercively capable of defeating outsider movements also enhance their likelihood of remaining intact following a regime transition, which I refer to as their posttransition fate.

The main component of formalizing coercive capabilities is straightforward: a competent military is endowed with greater coercive strength than a personalist military, $ {\theta}_{\mathrm{comp}}>{\theta}_{\mathrm{pers}} $ . Yet closing out the model requires imposing several additional, intuitive assumptions about precisely how coercive strength affects the probability of winning. The probability that coercion succeeds strictly increases in the military’s coercive endowment and strictly decreases in the outsider’s strength: $ \frac{\partial p}{\partial {\theta}_{\mathrm{mil}}}>0 $ and $ \frac{\partial p}{\partial {\theta}_{\mathrm{out}}}<0 $ .Footnote ¹³ Additionally, a stronger outsider threat amplifies the advantage of a more-capable state military, $ \frac{\partial^2p}{\partial {\theta}_{\mathrm{mil}}\partial {\theta}_{\mathrm{out}}}>0\hskip0.1em $ . Without this assumption, the ruler would face no incentive even in principle to turn to a more competent military when facing a strong outsider threat. Thus, this assumption incorporates a core premise of existing arguments. Finally, to eliminate substantively uninteresting corner solutions, I impose boundary conditions for very weak and very strong outsider threats. If the outsider is perfectly weak, then either type of state military defeats it for sure upon acting loyally. Formally, the lower-bound condition at $ {\theta}_{\mathrm{out}}=0 $ is $ p\left({\theta}_{\mathrm{mil}},0\right)=1 $ for any $ {\theta}_{\mathrm{mil}}>0 $ . Additionally, either type of state military retains some prospect for defeating a very strong outsider, perhaps because of inherent defensive advantages to guarding the capital. Formally, the upper bounds at $ {\theta}_{\mathrm{out}}\to \infty $ are $ 0<{p}_{\mathrm{pers}}^{\infty }<{p}_{\mathrm{comp}}^{\infty }<1 $ and $ {\alpha}^{\infty }>0 $ .Footnote ¹⁴ Figure 2 depicts a functional form for $ p\left(\cdot \right) $ that satisfies these assumptions.

Figure 2. Probability of Military Loyalty Resulting in Outsider Defeat

Note: Parameter values: $ p\left({\theta}_{\mathrm{mil}},{\theta}_{\mathrm{out}}\right)=\frac{1+{\theta}_{\mathrm{mil}}\cdot {\theta}_{\mathrm{out}}}{1+{\theta}_{\mathrm{out}}} $ , $ {\theta}_{\mathrm{comp}}=0.5 $ , $ {\theta}_{\mathrm{pers}}=0.05 $ .

The second component of the core trade-off is that a competent military anticipates a better posttransition fate than a personalist military. Following a regime transition in which leaders of the outsider movement gain control of the government, the personalist military consumes 0. By contrast, the competent military consumes $ {\pi}_{\mathrm{trans}}>0 $ upon defecting, and a fraction $ \gamma $ of this amount if they attempt but fail to prevent outsider rule, either by acting loyally or staging a coup.Footnote ¹⁵

The following two restrictions on $ {\pi}_{\mathrm{trans}} $ make the problem strategically interesting by ensuring that the competent military does not always prefer defecting over their other two options. First, the competent military prefers the incumbent regime over mass rule if Nature draws the highest possible valuation of the incumbent regime, denoted above as $ {\pi}_{\mathrm{sq}}^{\mathrm{max}} $ . Second, the competent military prefers to live under a military dictatorship (which yields consumption of 1) to living under mass rule. These two assumptions are formalized as $ {\pi}_{\mathrm{trans}}<\min \left\{{\pi}_{\mathrm{sq}}^{\mathrm{max}},1\right\} $ .

I intentionally omit from the model two other possible differences between competent and personalist militaries that would obscure the effect of divergent posttransition fates. First, both militaries gain the same consumption if the incumbent regime survives, $ {\pi}_{\mathrm{sq}}\hskip-0.2em $ . Thus, I do not assume that members of personalist militaries necessarily exhibit high intrinsic affinity for the ruler. Indeed, as shown below, they sometimes stage coups in equilibrium. Second, I do not assume that personalist militaries face greater hurdles to overthrowing the ruler, as both militaries topple the ruler for sure if they stage a coup. Instead, as I show in the analysis, the main difference between the competent and personalist military is that the latter anticipates a poor posttransition fate. Circumscribing their alternatives to supporting the ruler provides strategic microfoundations for the high likelihood with which the personalist military exhibits loyalty toward the ruler. This approach contrasts with less strategically interesting mechanisms: personalist militaries do not want to, or cannot, remove the ruler.

Simplifying Assumptions and Extensions

In the baseline model, the ruler makes a binary choice about how to organize the conventional military in anticipation of a specific and perfectly known outsider threat who is represented by a Nature move. In the appendix, I present two extensions that preserve the same core trade-off when relaxing each of these assumptions.

In the first extension (Appendix A.3), the ruler makes a continuous choice over how to allocate resources across two security units: the conventional military and a personalist paramilitary force. The ruler does not know exactly what type of outsider threat will arise and chooses which security unit to deploy only after learning this information. I also demonstrate the importance of fiscal health, which enables the ruler to better hedge their bets by allocating more resources for each unit.

In the second extension (Appendix A.4), I model the masses as a strategic actor who chooses whether to mobilize. Here I interpret the coercive agent as a secret police unit that uses repression to prevent a strategic mass actor from mobilizing, as opposed to using the conventional military to react to an existing mass threat.

Each extension also more explicitly expresses that the decisions in my model occur at different times in the real world. Rulers cannot instantaneously reorganize their coercive apparatus. Thus, their strategic choice reflects their expectations about future outsider threats, although in cases of long-running insurgencies, rulers can over time reorganize the coercive apparatus in response to an already-formed threat. Empirically, Geddes, Wright, and Frantz (Reference Geddes, Wright and Frantz2018, 85–9) show that dictators most frequently reshape their coercive apparatus (e.g., establishing personal control over promotions, creating a separate paramilitary) early in their tenures. However, rulers retain agency to make subsequent modifications if the dominant perceived threat changes over time (Greitens Reference Greitens2016). In the conclusion, I discuss impediments for rulers to create their preferred type of coercive apparatus.

Isolating the Outsider Threat

Existing theories of the guardianship dilemma focus on how the fear of a coup encourages many dictators to sacrifice competence. I instead show that competence can pose problems for the dictator even when facing an outsider threat only. I first analyze the model without the coup option, thus isolating the military’s decision between loyalty and defection.Footnote ¹⁶ A coercively strong outsider indeed increases the ruler’s desire for a more competent military. This motivation is consistent with the conventional wisdom. However, if the competent military has a favorable posttransition fate, then they are unreliable. Their endowed coercive advantage is irrelevant because they are unmotivated to save the incumbent regime.

The dictator’s objective is to maximize the probability of regime survival. Absent a threat of insider removal by the military, this is equivalent to maximizing the probability of defeating the outsider threat. This probability depends not only on the military’s coercive capacity but also on their incentives to act loyally. Loyalty is guaranteed from the personalist military, whose alternative is to defect and consume 0. By contrast, the competent military gains strictly positive consumption upon the regime falling. Consequently, the competent military attempts to save the regime if and only if Nature draws a sufficiently high valuation of the status quo, $ {\pi}_{\mathrm{sq}} $ :

(1) $$ {\displaystyle \begin{array}{l}\underset{\mathrm{Loyalty}}{\underbrace{p_{\mathrm{comp}}\cdot {\pi}_{\mathrm{sq}}+\left(1-{p}_{\mathrm{comp}}\right)\cdot \gamma \cdot {\pi}_{\mathrm{trans}}}}\hskip1em \ge \underset{\mathrm{Defect}}{\underbrace{\pi_{\mathrm{trans}}}}\\ {}\hskip1em \Rightarrow {\pi}_{\mathrm{sq}}\ge {\tilde{\pi}}_{\mathrm{sq}}^{\mathrm{def}}\equiv {\pi}_{\mathrm{trans}}\cdot \left[\left(1-\gamma \right)\cdot \frac{1}{p_{\mathrm{comp}}}+\gamma \right].\end{array}} $$

The incentive-compatibility constraint for the ruler to choose a competent military is

(2) $$ \underset{\Pr \left(\mathrm{loyalty}>\mathrm{defect}\right)}{\underbrace{\left[1-F\left({\tilde{\pi}}_{\mathrm{sq}}^{\mathrm{def}}\right)\right]}}\cdot {p}_{\mathrm{comp}}\ge {p}_{\mathrm{pers}}, $$

and $ F\left(\cdot \right) $ incorporates the probability draw for $ {\pi}_{\mathrm{sq}} $ .

Figure 3 provides visual intuition for the ensuing proposition. The figure is a region plot with outsider-threat strength $ {\theta}_{\mathrm{out}} $ on the x-axis and the competent military’s posttransition fate $ {\pi}_{\mathrm{trans}} $ on the y-axis.

Figure 3. Optimal Military Organization: Outsider Threat Only

Note: Parameter values: $ p\left({\theta}_{\mathrm{mil}},{\theta}_{\mathrm{out}}\right)=\frac{1+{\theta}_{\mathrm{mil}}\cdot {\theta}_{\mathrm{out}}}{1+{\theta}_{\mathrm{out}}} $ , $ {\theta}_{\mathrm{comp}}=0.3 $ , $ {\theta}_{\mathrm{pers}}=0.2 $ , $ {\pi}_{\mathrm{sq}}^{\mathrm{max}}=6 $ , $ \gamma =0.3 $ , $ {\pi}_{\mathrm{trans}}=0.5 $ .

Two distinct factors contribute to the conditions under which the personalist military is optimal, expressed by the white region in the figure. First, the competent military is unreliable if their anticipated posttransition fate is favorable, as in the top part of the figure. Only for particularly high draws of $ {\pi}_{\mathrm{sq}} $ is the competent military willing to exercise repression, given high $ {\pi}_{\mathrm{trans}} $ and their desire to not diminish that consumption amount if repression fails to prevent a regime transition. For high enough $ {\pi}_{\mathrm{trans}} $ , this effect swamps their endowed coercive advantage—even if $ {\theta}_{\mathrm{out}} $ is arbitrarily large. Latent competence is irrelevant from the ruler’s perspective if the military is unlikely to use it to save the regime. This highlights the importance of modeling repression as a strategic choice for the military rather than assuming compliance with repression orders.

Second, the competent military is unnecessary if the outsider threat is weak. In the left part of the figure, the gap between $ {p}_{\mathrm{comp}} $ and $ {p}_{\mathrm{pers}} $ is small because either type of military can easily defeat an outsider with low $ {\theta}_{\mathrm{out}} $ (see Figure 2). Thus, even if $ {\pi}_{\mathrm{trans}} $ is low—which enhances the competent military’s incentives to exercise repression—an even smaller difference in the probabilities of winning overshadows this effect.

The competent military maximizes the probability of defeating the outsider threat if and only if their posttransition fate is unfavorable and the outsider threat is severe, shown in the gray region. High $ {\theta}_{\mathrm{out}} $ yields a large latent coercive advantage for the competent military,Footnote ¹⁷ and low $ {\pi}_{\mathrm{trans}} $ engenders a high probability of acting loyally. This logic also explains why lower $ {\pi}_{\mathrm{trans}} $ increases the range of parameter values at which $ {\theta}_{\mathrm{out}} $ is large enough for the ruler to prefer the competent military. Proposition 1 presents the accompanying subgame perfect Nash equilibrium strategy profile.

Adding Insider Threats

Allowing the possibility of an insider coup by the military does not qualitatively change the ruler’s calculus. Footnote ¹⁸ This finding departs from existing theories in which the primary tension faced by the ruler is between choosing a personalist military to guard against coups or a competent military to guard against outsider threats. As in the preceding analysis, when I isolated the outsider threat, the ruler prioritizes competence if and only if the outsider threat is strong and the competent military anticipates a bad posttransition fate. Defection and coups are two variants of disloyalty, and a better posttransition option raises the attractiveness of either alternative relative to acting loyally.

The ruler’s objective is, as before, to maximize the probability of survival. Now survival additionally requires the military to not stage a coup. The availability of this strategic option changes the calculus of each military actor. Unlike before, the personalist military is not guaranteed to act loyally. For any draw $ {\pi}_{\mathrm{sq}}<1 $ , their best possible outcome is to establish a military dictatorship, which yields consumption of 1. Yet coups weaken the center and elevate the likelihood of outsider takeover relative to acting loyally, captured by $ \alpha <1 $ . Thus, coups are riskier than acting loyally. The incentive-compatibility constraint for the personalist military to act loyally is

(3) $$ \underset{\mathrm{Loyalty}}{\underbrace{p_{\mathrm{pers}}\cdot {\pi}_{\mathrm{sq}}}}\ge \underset{\mathrm{Coup}}{\underbrace{\alpha \cdot {p}_{\mathrm{pers}}\cdot 1}}\Rightarrow {\pi}_{\mathrm{sq}}\ge \alpha . $$

The competent military has three strategically relevant options, which complicates their optimal decision. The incentive-compatibility constraint to act loyally is

(4) $$ {\displaystyle \begin{array}{l}\underset{\mathrm{Loyalty}}{\underbrace{p_{\mathrm{comp}}\cdot {\pi}_{\mathrm{sq}}+\left(1-{p}_{\mathrm{comp}}\right)\cdot \gamma \cdot {\pi}_{\mathrm{trans}}}}\\ {}\hskip.75em \ge {\displaystyle \begin{array}{l}\max \Big\{\underset{\mathrm{Defect}}{\underbrace{\pi_{\mathrm{trans}}}},\underset{\mathrm{Coup}}{\underbrace{\alpha \cdot {p}_{\mathrm{comp}}\cdot 1+\left(1-\alpha \cdot {p}_{\mathrm{comp}}\right)\cdot \gamma \cdot {\pi}_{\mathrm{trans}}}}\Big\}.\end{array}}\end{array}} $$

I solve for the equilibrium probability of loyalty in two steps. First, I evaluate bilateral comparisons between loyalty and each disloyalty option. I already compared loyalty with defection and derived a threshold $ {\tilde{\pi}}_{\mathrm{sq}}^{\mathrm{def}} $ in Equation 1. Regarding loyalty versus coup, the threshold value of $ {\pi}_{\mathrm{sq}} $ that induces loyalty is

(5) $$ {\pi}_{\mathrm{sq}}\ge {\tilde{\pi}}_{\mathrm{sq}}^{\mathrm{coup}}\equiv \alpha +\left(1-\alpha \right)\cdot \gamma \cdot {\pi}_{\mathrm{trans}}. $$

This inequality shows that higher $ {\pi}_{\mathrm{trans}} $ increases the competent military’s preference for a coup relative to acting loyally. Regime transition is more likely to occur following a coup than if the military loyally guards the regime (because $ \alpha <1 $ ). Higher $ {\pi}_{\mathrm{trans}} $ makes this discrepancy in the probability of a bad outcome less relevant by improving the bad outcome. The competent military is less averse to coups because higher $ {\pi}_{\mathrm{trans}} $ increases their consumption upon failing to prevent outsider rule.

The second step for establishing the probability with which the competent military acts loyally is to compare their two disloyalty options. I show that their most-preferred disloyalty option is a coup if the outsider threat is weak, and defection if strong. The competent military fares better under a military dictatorship (consumption of 1) than following a regime transition ( $ {\pi}_{\mathrm{trans}}<1 $ ). Yet coups are risky. Consumption is $ \gamma \cdot {\pi}_{\mathrm{trans}} $ following a failed coup, given the penalty of magnitude $ 1-\gamma $ that the masses impose against the military for exercising repression. A stronger outsider threat causes the competent military to place more weight on the failed-coup outcome, which increases their preference for defection relative to staging a coup. The following formalizes a threshold $ {\tilde{\theta}}_{\mathrm{out}}^{\mathrm{dis}} $ such that the binding constraint is a coup if $ {\theta}_{\mathrm{out}}<{\tilde{\theta}}_{\mathrm{out}}^{\mathrm{dis}} $ and defection if $ {\theta}_{\mathrm{out}}\ge {\tilde{\theta}}_{\mathrm{out}}^{\mathrm{dis}} $ . Appendix Lemma A.1 proves that this threshold is unique and characterizes its bounds.

(6) $$ \begin{array}{c}\underset{\mathrm{Coup}}{\underbrace{\alpha \left({\tilde{\theta}}_{\mathrm{out}}^{\mathrm{dis}}\right)\cdot p\left({\theta}_{\mathrm{comp}},{\tilde{\theta}}_{\mathrm{out}}^{\mathrm{dis}}\right)+\left[1-\alpha \left({\tilde{\theta}}_{\mathrm{out}}^{\mathrm{dis}}\right)\cdot p\left({\theta}_{\mathrm{comp}},{\tilde{\theta}}_{\mathrm{out}}^{\mathrm{dis}}\right)\right]\cdot \gamma \cdot {\pi}_{\mathrm{trans}}}}\\ {}=\underset{\mathrm{Defect}}{\underbrace{\pi_{\mathrm{trans}}}}.\end{array} $$

These steps imply that the exact form of the incentive-compatibility constraint for the ruler to choose a competent military depends on $ {\theta}_{\mathrm{out}} $ :

(7) $$ {\displaystyle \begin{array}{l}\underset{\Pr \left(\mathrm{loyalty}>\mathrm{coup}\right)}{\underbrace{\left[1-F\left({\tilde{\pi}}_{\mathrm{sq}}^{\mathrm{coup}}\right)\right]}}\cdot {p}_{\mathrm{comp}}\hskip.2em \ge {\displaystyle \begin{array}{l}\underset{\Pr \left(\mathrm{loyalty}>\mathrm{coup}\right)}{\underbrace{\left[1-F\left(\alpha \right)\right]}}\cdot {p}_{\mathrm{pers}}\\ {}\mathrm{if}\underset{\mathrm{Comp}.\ \mathrm{mil}.\mathrm{prefers}\ \mathrm{coup}}{\underbrace{\theta_{\mathrm{out}}<{\tilde{\theta}}_{\mathrm{out}}^{\mathrm{dis}}}}\end{array}}\\ {}\underset{\Pr \left(\mathrm{loyalty}>\mathrm{defect}\right)}{\underbrace{\left[1-F\left({\tilde{\pi}}_{\mathrm{sq}}^{\mathrm{def}}\right)\right]}}\cdot {p}_{\mathrm{comp}}\hskip.2em \ge {\displaystyle \begin{array}{l}\underset{\Pr \left(\mathrm{loyalty}>\mathrm{coup}\right)}{\underbrace{\left[1-F\left(\alpha \right)\right]}}\cdot {p}_{\mathrm{pers}}\\ {}\mathrm{if}\underset{\mathrm{Comp}.\ \mathrm{mil}.\mathrm{prefers}\ \mathrm{defect}}{\underbrace{\theta_{\mathrm{out}}\ge {\tilde{\theta}}_{\mathrm{out}}^{\mathrm{dis}}}}.\end{array}}\end{array}} $$

This expression resembles the incentive-compatibility constraint in Equation 2. Introducing a coup threat does not qualitatively alter the ruler’s calculus compared with facing an outsider threat only. The two primary effects that drive Proposition 1 are still at work. First, higher $ {\theta}_{\mathrm{out}} $ enhances the value-added of greater coercive capabilities. Second, lower $ {\pi}_{\mathrm{trans}} $ boosts the reliability of the competent military because a worse posttransition fate diminishes their valuation of either disloyalty option relative to acting loyally. Figure 4 is analogous to Figure 3. The only important difference is that there are two darker regions that each correspond with a specific disloyalty option. Proposition 2 formally characterizes the ruler’s optimal choice.Footnote ¹⁹

Figure 4. Optimal Military Organization: Dual Disloyalty Options

Note: Parameter values: Same as Figure 3, and $ \alpha \left({\theta}_{\mathrm{out}}\right)=\frac{0.3+0.1\cdot {\theta}_{\mathrm{out}}}{1+{\theta}_{\mathrm{out}}} $ .

Equilibrium Probability of a Coup

According to the canonical logic of the guardianship dilemma, the equilibrium probability of a coup should increase in the severity of the outsider threat. Why? A more competent military is needed to defeat a stronger outsider threat, but such militaries are also more prone to stage coups. Thus, the ruler tolerates a higher probability of insider removal to mitigate prospects for outsider overthrow.

I set up the model so that, all else equal, the competent military is indeed more prone to staging a coup. Compared with the personalist military, the competent military has a lower opportunity cost to staging a coup (relative to acting loyally). They consume $ \gamma \cdot {\pi}_{\mathrm{trans}} $ even if they fail to consolidate power, whereas the personalist military consumes 0. This logic yields Lemma 1.Footnote ²⁰

Yet my overall implications for the relationship between the severity of the outsider threat and the equilibrium probability of a coup depart from convention. My contrarian findings highlight a crucial difference between all-else-equal propositions and equilibrium relationships. In equilibrium, the competent military may be less likely than the personalist military to stage a coup because of selection and substitution effects.

Unfavorable posttransition fate. I partially recover conventional implications if the competent military’s posttransition consumption, $ {\pi}_{\mathrm{trans}} $ , is low enough. In this case, defection is strategically irrelevant for the competent military because they strictly prefer a coup over defecting.Footnote ²¹ Figure 5 depicts the relationship. The strength of the outsider threat, $ {\theta}_{\mathrm{out}} $ , is on the x-axis, and the equilibrium probability of a coup, which I denote as $ \Pr \left({\mathrm{coup}}^{\ast}\right) $ , is on the y-axis.Footnote ²² The ruler switches from a personalist to a competent military when the outsider threat is severe enough, $ {\theta}_{\mathrm{out}}={\hat{\theta}}_{\mathrm{out}}^{\mathrm{dual}} $ .Footnote ²³ This switch yields a discrete increase in $ \Pr \left({\mathrm{coup}}^{\ast}\right) $ .Footnote ²⁴ Thus, under conditions in which the competent military does not defect, my model recovers a central implication of the canonical guardianship dilemma logic.

Figure 5. Equilibrium Probability of a Coup: Unfavorable Posttransition Fate

Note: Solid segments of curves correspond with parameter values at which the ruler optimally chooses the specified type of military (black for competent military, gray for personalist). Pr(coup*) equals the piecewise function created by the solid segments of curves. Dashed segments correspond with off-the-equilibrium path outcomes. These express what the probability of a coup would be if the ruler chose their less-preferred type of military (at those parameter values). See Proposition 2 for $ {\hat{\theta}}_{\mathrm{out}}^{\mathrm{dual}} $ . The parameter values are the same as in Figure 4, while additionally setting $ {\pi}_{\mathrm{trans}}=0.12 $ . This value of $ {\pi}_{\mathrm{trans}} $ is marked on the y-axis of Figure 4, and the main insights from Figure 5 can be gleaned from moving horizontally across Figure 4 while fixed at this value of the y-axis. In Figure 5, the range of the x-axis is truncated compared with Figure 4 to highlight the discrete jump more clearly.

Yet even when the competent military does not defect, the conventional logic still requires modification. In equilibrium, the competent military does not necessarily pose a starker insider coup threat because of a selection effect driven by the following two elements. (1) The ruler prioritizes competence only if the outsider is sufficiently strong. (2) Strong outsider threats lower the propensity for either type of military to stage a coup by decreasing the probability that the military can consolidate power.Footnote ²⁵ For the parameter values in Figure 5, $ \Pr \left({\mathrm{coup}}^{\ast}\right) $ is higher at $ {\theta}_{\mathrm{out}}=0 $ , at which point the ruler chooses the personalist military, than at higher values (such as $ {\theta}_{\mathrm{out}}=0.2 $ ) for which the ruler prioritizes competence.

Favorable posttransition fate. The implications depart more starkly from conventional characterizations of the guardianship dilemma when the competent military’s posttransition consumption, $ {\pi}_{\mathrm{trans}} $ , is higher. I develop a substitution effect that explains why the competent military does not necessarily pose a greater coup threat than the personalist military (even after accounting for the selection effect just described).

I depict the relationship in Figure 6. This figure is identical to Figure 5 except $ {\pi}_{\mathrm{trans}} $ is fixed at a higher value in each panel. Two aspects of Panel A of Figure 6 are identical to Figure 5: The ruler switches from the personalist to the competent military for high enough $ {\theta}_{\mathrm{out}} $ , and $ \Pr \left({\mathrm{coup}}^{\ast}\right) $ exhibits a discrete positive jump at this point. The main difference is that at an even higher value of $ {\theta}_{\mathrm{out}} $ , the competent military’s preferred disloyalty option switches from coup to defect. Thus, defection substitutes from the competent military’s desire to stage a coup. This eliminates the dreaded insider threat stressed in existing theories of the guardianship dilemma, as shown by $ \Pr \left({\mathrm{coup}}^{\ast}\right) $ dropping to 0.

Figure 6. Equilibrium Probability of a Coup: Better Posttransition Fate

Note: See the note for Figure 5. The parameter values are the same as Figure 4, while additionally setting $ {\pi}_{\mathrm{trans}}=0.22 $ in Panel A and $ {\pi}_{\mathrm{trans}}=0.6 $ in Panel B. These values of $ {\pi}_{\mathrm{trans}} $ are marked on the y-axis of Figure 4, and the main insights from Figure 6 can be gleaned from moving horizontally across Figure 4 while fixed at either of these values of the y-axis.

In Panel B, an even better posttransition fate makes the competent military so unreliable that the ruler prefers the personalist military against an arbitrarily strong outsider threat. Consequently, the equilibrium probability of a coup strictly decreases in $ {\theta}_{\mathrm{out}} $ , which is the opposite relationship from conventional theories.Footnote ²⁶ Yet counterfactually, if the ruler chose the competent military, the probability of a coup would be 0 because of the defection-for-coup substitution effect. Proposition 3 formalizes the threshold values of $ {\pi}_{\mathrm{trans}} $ that distinguish the cases depicted in Figures 5 and 6.

Reframing the Guardianship Dilemma

Existing theories of the guardianship dilemma focus on the fear of authoritarian guards staging a coup. I reframe the guardianship dilemma around a different idea. The fundamental problem a competent military poses for a ruler is life beyond the incumbent regime. Paradoxically, a favorable posttransition fate eliminates the coup threat posed by a competent military, but it also makes the dictator worse off. The competent military substitutes into an even better disloyalty option: defecting and acquiescing to outsider rule. The same effect that makes the competent military less of an insider threat also undercuts their reliability for combating rebellions and popular uprisings. Conversely, an unfavorable posttransition fate causes the competent military to prefer coups over defection. Despite creating a threat of insider removal, the dictator prefers this scenario because they can count on the competent military to exercise coercion against outsider movements.

Formally, the ruler’s prospects for survival (weakly) decrease in $ {\pi}_{\mathrm{trans}} $ . By contrast, for some parameter values, the ruler’s survival prospects decrease even as the equilibrium probability of a coup decreases. I illustrate these equilibrium implications in Figure 7, in which posttransition consumption for the competent military is on the x-axis. Panel A presents the ruler’s probability of survival, and Panel B presents the probability of a coup. For low values of $ {\pi}_{\mathrm{trans}} $ , the relationship aligns with convention: the probability of ruler survival decreases while that of a coup increases. However, for intermediate values of $ {\pi}_{\mathrm{trans}} $ , the relationship inverts. The probability that the ruler survives continues to drop in $ {\pi}_{\mathrm{trans}} $ despite zero risk of a coup. The ruler optimally chooses a competent military, but they pose no insider threat because defection substitutes for coups. Finally, for higher values of $ {\pi}_{\mathrm{trans}} $ , the competent military becomes so unreliable that the ruler switches to the personalist military, who pose a moderate coup threat but will not defect. Appendix Proposition A.1 provides a supporting formal statement.

Figure 7. How Posttransition Fate Influences Equilibrium Outcomes

Note: Same parameter values as previous figures, plus $ {\theta}_{\mathrm{out}}=0.3 $ . See Appendix Equations A.21 and A.22 for $ {\tilde{\pi}}_{\mathrm{trans}}^{\mathrm{dis}} $ and $ {\hat{\pi}}_{\mathrm{trans}}^{\mathrm{def}} $ , respectively.

Radical Redistributive Threats

In the model, the competent military’s behavior depends on the goals of the outsider movement they confront. Dictators can, paradoxically, benefit when they face outsider movements that espouse radical redistributive intentions. Such mass organizations seek to transform the composition of the elite class and perhaps the entire social structure. Even a competent military fears their fate if a radical movement succeeds, yielding low $ {\pi}_{\mathrm{trans}} $ . Here I operationalize radical and nonradical threats, summarize the empirical prevalence of both types, and contrast strategies pursued by rulers in Rwanda and Kenya in response to divergent types of outsider threats.

Summary statistics. Existing research highlights various types of outsider movements that seek radical redistributive outcomes. Marxist insurgents seek economic redistribution. For example, the Chinese Communist party implemented a massive land reform during and after its struggle to capture power in 1949 to “destroy the gentry-landlord class (and thus eliminate a potential counterrevolutionary threat), establish Communist political power within the villages, and thus promote the building of a centralized state with firm administrative control over the countryside” (Meisner Reference Meisner1999, 92). In other cases, rebels seek radical redistribution along identity lines to reverse horizontal inequalities. This includes rebels that seek to capture the state and displace the ruling ethnic group with their own or that intend to create a regime based on violent interpretations of Islamic principles.

Figure 8 shows that dictators have frequently confronted radical outsider threats during the Cold War (1945–91) and afterward (1992–2015). The first row is any center-seeking civil war in which rebels seek to capture the capital city. Successful insurgencies often replace the state military with the rebel military, although not all such movements espouse radical objectives and gravely threaten the state military (e.g., the Chad example discussed earlier in the article).Footnote ²⁸ The next three rows disaggregate center-seeking rebel groups that typically pose unambiguously radical threats: Marxist, violent Islamist, or ethnic objectives. Although Marxist movements largely ended with the fall of the Soviet Union, Islamist rebels and ethnic rebels have each gained in frequency since the Cold War ended.

Figure 8. Outsider Threats in Dictatorships

Note: Each row represents the fraction of country-years with the specified event. Data described in Appendix A.5.

Mass organizations with radical redistributive goals contrast with nonviolent and pro-democracy movements that seek to oust the existing regime but, typically, not to overturn the entire social structure (Brancati Reference Brancati2016). Recently, nonviolent movements have increased in prevalence, as shown in the last row of Figure 8. My model highlights that nonradical movements pose a grave danger to authoritarian regimes because they reduce incentives for a competent military to act loyally. The recent proliferation of elections presents a similar difficulty for authoritarian rulers (Levitsky and Way Reference Levitsky and Way2010). Incumbents often deploy the security forces before, during, and after election day to prevent opposition victory. However, a broad-based military may be less willing to save the regime against a challenger operating through institutionalized channels and, often, backed by Western monitoring.

The model explains why rulers should craft more personally oriented units in response to nonradical outsider threats. Speculatively, although consistent with this expectation, the frequency of personalist characteristics in militaries has increased since the Cold War ended. I show this in Figure 9 by presenting data from Geddes, Wright, and Frantz (Reference Geddes, Wright and Frantz2018) on three aspects of military personalism: control, paramilitaries, and promotion. The rise in military personalization since the Cold War ended is striking in contrast to the general trend of greater institutionalization within dictatorships over this period (Meng Reference Meng2020). Thus, Figure 9 highlights a pattern that would be fruitful for future research to analyze.

Figure 9. Military Personalism in Dictatorships

Note: Each row represents the fraction of country-years with the specified trait. Data described in Appendix A.5.

Empirical cases. Rwanda provides an illustrative case of a regime responding to a radical outsider threat by creating a socially inclusive and professional military. In 1995, the Tutsi-dominated Rwandan Patriotic Front (RPF) overthrew the government and replaced the state army with their armed wing, the Rwandan Patriotic Army (RPA). The RPF contemplated whether to keep the military exclusive to Tutsis, who comprised about 15% of the population, or to expand by incorporating Hutus. Rwanda’s long history of racial tensions between Hutus and Tutsis rules out many possible alternative explanations for why a ruler would broaden the ethnic basis of their military. After the Hutu Revolution of 1959 terminated the historical Tutsi monarchy, Hutus monopolized political and military positions from independence through the mid-1990s. Prior to takeover by the RPF, a negotiated settlement failed that included a provision for military integration. This spurred the Rwandan genocide against Tutsis in 1994, and then the invasion by the RPF.

Despite bloody ethnic antagonisms, the RPF immediately sought to make the new state army socially inclusive. During the RPF’s campaign to seize power, many Rwandans with extremist beliefs about Hutu superiority fled to neighboring Zaire and posed a strong radical threat to the new regime. Acknowledging this threat, “the RPF regime sought to ensure the security and defense of the country by forming a coherent national defense force, and it thus began the process of converting the RPA from a guerrilla army into a larger and more conventional force that could defend the country.” Incorporating numerous Hutu soldiers from the ex-state army was “[o]bviously a big risk.” However, regime elites deemed this move necessary to counter the large and radical outsider threat. These reforms resulted in “one of the most capable militaries in Africa” (Burgess Reference Burgess and Licklider2014, 92, 97).

Kenya provides an illustrative case of a regime responding to rising nonradical outsider threats by making its coercive apparatus more ethnically exclusive. Following the loss of unconditional aid from the United States and a failed crackdown of a peaceful pro-democracy movement in 1990–91, the incumbent ruler Daniel arap Moi (an ethnic Kalenjin) was forced to concede multiparty elections in 1992. At this point, “viable opposition campaigns” became the main threat to the regime, as opposed to a threat of a coup (Hassan Reference Hassan2020, 97).Footnote ²⁹ This case helps to isolate the strategic reaction to nonradical outsider threats. The nature of the main threat to the regime changed because an exogenous shock, the Cold War ending, caused Kenya’s primary Western benefactor to lower its tolerance of autocrats.

The regime responded to new outside challenges by recruiting (along ethnic lines) actors outside the conventional army to repress opponents: “‘warriors’ of Kalenjin and Maasai ethnicity, groups strongly represented in the ruling party, and more recently KANU ‘youthwingers’ provided another mechanism of control by the state” (Kirschke Reference Kirschke2000, 398; see also Levitsky and Way Reference Levitsky and Way2010, 267–9). During this period, Geddes, Wright, and Frantz (Reference Geddes, Wright and Frantz2018) switch their coding of military promotions in Kenya from predominantly based on merit to predominantly based on ethnic ties. Between 1988 and 1993, arap Moi reduced the number of rival Kikuyu and Luo elites—the ethnic basis of the main opposition parties—in the cabinet from thirteen to two (Hassan Reference Hassan2020, 100).

Weak Outsider Threats

When outsider threats are weak (low $ {\theta}_{\mathrm{out}} $ ), dictators do not need a military with high coercive capacity. In this circumstance, I anticipate that rulers will prioritize soldiers with a poor posttransition fate, which induces them to shoot on command. This differs from the mechanism posited in existing theories of the guardianship dilemma. Existing theories also expect that rulers will react to weak outsider threats with personalist military recruitment, specifically because they fear coup attempts by a more competent military. Empirically, it is likely that prospects for defection and for coups each influence a dictator’s calculus. To isolate empirical support for the preventing-defection mechanism, I select a case in which a coup was essentially impossible: European colonies in Africa. The mechanism in my model provides a strategic basis for racist “martial race” theories of colonial military recruitment.

During the interwar period, European colonial rulers feared neither mass outsider movements nor insider coups. By this time, European powers had successfully repressed major precolonial states that resisted colonial imposition and had put down early antitax revolts, and almost no internal wars occurred in African colonies between 1919 and 1939. European powers jointly agreed to fixed borders and to not fight wars over their African territories, which minimized outsider threats from European challengers. Colonial states also had external security guarantees from the metropole if a widespread rebellion emerged or a coup attempt occurred. Coup attempts were also extremely unlikely because Europeans dominated the officer corps of colonial militaries.

Consequently, European colonial officials primarily sought to select rank-and-file soldiers that would loyally follow commands to repress small-scale disturbances. Colonial officials anticipated that the greatest need for force would be in the capital city. They often turned to groups of people in the periphery that lacked ethnic ties to residents of the capital, thus anticipating a poor posttransition fate if groups from the colonial center gained the upper hand. Frederick Lugard, an influential and notorious colonial administrator, wrote, “Where a handful of white men are engaged in the difficult task of introducing peace and good government … the chief danger … lies in possible disaffection among the troops.” He favored “battalions or wings of battalions, composed of races which have no affinities with the population of the region in which they are serving, and even the introduction of an alien battalion may be a wise precaution” (Lugard Reference Lugard1922, 577). This logic led Lugard and other administrators to create myths of “martial” prowess for peripheral groups that they prioritized in the colonial military.